Pseudo Camera 10K
Dataset containing 10,000 high-resolution images from professional photographers, with automatic captions generated by CogVLM. This corpus is varied, with numerous portraits and annotated group scenes.
10,000 high-resolution images, automatic caption annotations, JPEG/PNG formats
Creative Commons (exact type to be specified)
Description
The dataset Pseudo Camera 10K offers 10,000 high-quality images, resized with particular care to limit artifacts. The images are annotated with captions generated automatically by CogVLM, which allows immediate use to train vision-linguistic models.
What is this dataset for?
- Train vision models with automatic text annotations
- Improve the recognition and description of complex images, including portraits and groups
- Exploring the impact of natural noise and imperfect annotations on learning
Can it be enriched or improved?
Yes, legends can be corrected manually to improve accuracy. Metadata about cultural diversity or photo contexts can also be added.
🔎 In summary
🧠 Recommended for
- Vision researchers
- Captioning projects
- Data annotation teams
🔧 Compatible tools
- CogVLM
- PyTorch
- TensorFlow
- Linguistic vision frameworks
💡 Tip
Correct or enrich legends to improve the quality of the trained model.
Frequently Asked Questions
Are the automatic captions completely accurate?
No, they are close to 100% but may contain errors and require manual verification for critical uses.
Is this dataset suitable for training an image captioning model?
Yes, it is designed to provide a varied corpus with automatic captions, useful for supervised captioning.
What is the resolution of the images?
The images have a minimum resolution with the smallest edge at 1024 pixels, guaranteeing good visual quality.
